[Kernel] Porting the TRTLLM minimax_allreduce_rms kernels (#37045)

Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>

.buildkite/test_areas/kernels.yaml

@@ -20,7 +20,20 @@ steps:
   - tests/kernels/core
   - tests/kernels/test_concat_mla_q.py
   commands:
-    - pytest -v -s kernels/core kernels/test_concat_mla_q.py
+    - pytest -v -s kernels/core --ignore=kernels/core/test_minimax_reduce_rms.py  kernels/test_concat_mla_q.py
+
+- label: Kernels MiniMax Reduce RMS Test (2 GPUs)
+  timeout_in_minutes: 15
+  num_devices: 2
+  device: h100
+  source_file_dependencies:
+  - csrc/minimax_reduce_rms_kernel.cu
+  - csrc/minimax_reduce_rms_kernel.h
+  - vllm/model_executor/layers/mamba/linear_attn.py
+  - vllm/model_executor/layers/mamba/lamport_workspace.py
+  - tests/kernels/core/test_minimax_reduce_rms.py
+  commands:
+    - pytest -v -s kernels/core/test_minimax_reduce_rms.py
 
 - label: Kernels Attention Test %N
   timeout_in_minutes: 35

CMakeLists.txt

@@ -307,6 +307,8 @@ set(VLLM_EXT_SRC
   "csrc/torch_bindings.cpp")
 
 if(VLLM_GPU_LANG STREQUAL "CUDA")
+  list(APPEND VLLM_EXT_SRC "csrc/minimax_reduce_rms_kernel.cu")
+
   SET(CUTLASS_ENABLE_HEADERS_ONLY ON CACHE BOOL "Enable only the header library")
 
   # Set CUTLASS_REVISION. Used for FetchContent. Also fixes some bogus messages when building.

csrc/minimax_reduce_rms_kernel.cu

+
+/*
+ * Copyright (c) 2026, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <cooperative_groups.h>
+#include <cuda_runtime.h>
+
+#include <torch/cuda.h>
+#include <ATen/cuda/CUDAContext.h>
+#include <c10/cuda/CUDAGuard.h>
+
+#include "cuda_compat.h"
+#include "cuda_utils.h"
+#include "core/registration.h"
+#include "minimax_reduce_rms_kernel.h"
+
+#include <algorithm>
+
+#define FINAL_MASK 0xffffffff
+#define MINIMAX_REDUCE_RMS_WARP_SIZE 32
+
+namespace vllm {
+namespace tensorrt_llm {
+
+template <int NRanks>
+struct LamportComm {
+  __device__ __forceinline__ LamportComm(void** workspace, int rank) {
+    counter_ptr = &reinterpret_cast<int*>(workspace[NRanks * 3])[0];
+    flag_ptr = &reinterpret_cast<int*>(workspace[NRanks * 3])[2];
+    clear_ptr = &reinterpret_cast<int64_t*>(workspace[NRanks * 3 + 1])[0];
+    flag_value = *flag_ptr;
+    auto comm_size = reinterpret_cast<int64_t*>(workspace[NRanks * 3 + 1])[1];
+    clear_size = *clear_ptr;
+    int data_offset = flag_value % 3;
+    int clear_offset = (flag_value + 2) % 3;
+    for (int r = 0; r < NRanks; ++r) {
+      data_bufs[r] = reinterpret_cast<uint8_t*>(workspace[2 * NRanks + r]) +
+                     data_offset * comm_size;
+    }
+    clear_buf = reinterpret_cast<uint8_t*>(workspace[2 * NRanks + rank]) +
+                clear_offset * comm_size;
+    __syncthreads();
+    if (threadIdx.x == 0) {
+      atomicAdd(counter_ptr, 1);
+    }
+  }
+
+  __device__ __forceinline__ void update(int64_t new_clear_size) {
+    if (blockIdx.x == 0 && threadIdx.x == 0) {
+      while (*reinterpret_cast<int volatile*>(counter_ptr) != gridDim.x) {
+      }
+      *flag_ptr = (flag_value + 1) % 3;
+      *clear_ptr = new_clear_size;
+      *counter_ptr = 0;
+    }
+  }
+
+  int* counter_ptr;
+  int* flag_ptr;
+  int64_t* clear_ptr;
+  uint8_t* data_bufs[NRanks];
+  uint8_t* clear_buf;
+  int64_t clear_size;
+  int flag_value;
+};
+
+__device__ __forceinline__ bool is_neg_zero(float v) {
+  return *reinterpret_cast<uint32_t*>(&v) == 0x80000000;
+}
+
+__device__ __forceinline__ bool is_neg_zero(float4 v) {
+  return is_neg_zero(v.x) || is_neg_zero(v.y) || is_neg_zero(v.z) ||
+         is_neg_zero(v.w);
+}
+
+__device__ __forceinline__ float4 get_neg_zero() {
+  float4 vec;
+#pragma unroll
+  for (int i = 0; i < 4; ++i) {
+    reinterpret_cast<uint32_t*>(&vec)[i] = 0x80000000;
+  }
+  return vec;
+}
+
+template <int Dim>
+__device__ __forceinline__ float rms_rsqrt(float& v, float eps) {
+  constexpr float kInvDim = 1.0F / static_cast<float>(Dim);
+  v = rsqrtf((v * kInvDim) + eps);
+  return v;
+}
+
+template <int Dim>
+__device__ __forceinline__ float4 rms_rsqrt(float4& v, float eps) {
+  constexpr float kInvDim = 1.0F / static_cast<float>(Dim);
+  v.x = rsqrtf((v.x * kInvDim) + eps);
+  v.y = rsqrtf((v.y * kInvDim) + eps);
+  v.z = rsqrtf((v.z * kInvDim) + eps);
+  v.w = rsqrtf((v.w * kInvDim) + eps);
+  return v;
+}
+__device__ __forceinline__ float4 ld_global_volatile(float4* addr) {
+  float4 val;
+  asm volatile("ld.volatile.global.v4.f32 {%0, %1, %2, %3}, [%4];"
+               : "=f"(val.x), "=f"(val.y), "=f"(val.z), "=f"(val.w)
+               : "l"(addr));
+  return val;
+}
+
+__device__ __forceinline__ float ld_global_volatile(float* addr) {
+  float val;
+  asm volatile("ld.volatile.global.f32 %0, [%1];" : "=f"(val) : "l"(addr));
+  return val;
+}
+
+// Used by the scalar (non-float4) kernel only
+template <typename T, int NUM>
+__inline__ __device__ T warpReduceSumV2(T* val) {
+#pragma unroll
+  for (int i = 0; i < NUM; i++) {
+#pragma unroll
+    for (int mask = 16; mask > 0; mask >>= 1)
+      val[i] += __shfl_xor_sync(FINAL_MASK, val[i], mask, 32);
+  }
+  return (T)(0.0f);
+}
+
+template <typename T, int NUM>
+__inline__ __device__ T blockReduceSumV2(T* val) {
+  static __shared__ T shared[NUM][33];
+  int lane = threadIdx.x & 0x1f;
+  int wid = threadIdx.x >> 5;
+
+  warpReduceSumV2<T, NUM>(val);
+
+  if (lane == 0) {
+#pragma unroll
+    for (int i = 0; i < NUM; i++) {
+      shared[i][wid] = val[i];
+    }
+  }
+
+  __syncthreads();
+
+  bool is_mask = threadIdx.x < (blockDim.x / 32.f);
+#pragma unroll
+  for (int i = 0; i < NUM; i++) {
+    val[i] = is_mask ? shared[i][lane] : (T)(0.0f);
+  }
+  warpReduceSumV2<T, NUM>(val);
+  return (T)0.0f;
+}
+
+// for float4 version
+template <uint32_t kNumThreads, typename T, int ArraySize = 4>
+__device__ __forceinline__ void local_warp_reduce_sum_array(
+    T* value_ptr, uint32_t active_mask = 0xffffffffu) {
+  static_assert(kNumThreads >= 1 &&
+                kNumThreads <= MINIMAX_REDUCE_RMS_WARP_SIZE);
+#pragma unroll
+  for (int i = 0; i < ArraySize; ++i) {
+#pragma unroll
+    for (int mask = kNumThreads / 2; mask > 0; mask >>= 1) {
+      value_ptr[i] += __shfl_xor_sync(active_mask, value_ptr[i], mask,
+                                      MINIMAX_REDUCE_RMS_WARP_SIZE);
+    }
+  }
+}
+
+constexpr int next_pow2(int val) {
+  int result = 1;
+  while (result < val) {
+    result <<= 1;
+  }
+  return result;
+}
+
+// ---------------------------------------------------------------------------
+
+template <typename DType>
+class IndexHelper {
+ public:
+  __device__ __forceinline__ IndexHelper(MiniMaxReduceRMSParams const& params) {
+#if (defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 900))
+    namespace cg = cooperative_groups;
+    cg::cluster_group cluster = cg::this_cluster();
+    cg::grid_group grid = cg::this_grid();
+    token_id = grid.cluster_rank();
+    access_id_in_token = cluster.thread_rank();
+    token_stride = grid.num_clusters();
+#else
+    token_id = blockIdx.x;
+    access_id_in_token = threadIdx.x;
+    token_stride = gridDim.x;
+#endif
+    access_id = token_id * params.hidden_dim / kElemsPerAccess<DType> +
+                access_id_in_token;
+    access_stride = token_stride * params.hidden_dim / kElemsPerAccess<DType>;
+    tot_access = params.size_q / kElemsPerAccess<DType>;
+  }
+
+  int token_id;
+  int access_id_in_token;
+  int token_stride;
+  int access_id;
+  int access_stride;
+  int tot_access;
+};
+
+/**
+* this kernel is used to for minimax attention module
+* input tensor [total_tokens, hidden_dim / tp_size], fp32
+* rms weight [hidden_dim / tp_size], bf16
+step 1: reduce from single rank to get the variance sum (reduce(input^2,
+dim=-1)) step 2: reduce from all ranks to get the variance sum
+(all_reduce(variance_sum)) step 3: calculate the rms norm (input *
+rsqrt(variance + eps)) in this case, max hidden_dim is 6144 (float data), for
+each token, we only need 6144 / 4 / tp_size = (1536 / tp_size) threads so we can
+assume cluster size is 1 (tp_size >= 2)
+ */
+template <typename DType, int NRanks>
+__global__ void __launch_bounds__(1024)
+    minimax_reduce_rms_kernel_lamport(MiniMaxReduceRMSParams params) {
+  IndexHelper<DType> index_helper(params);
+  int token_id = index_helper.token_id;
+  int access_id_in_token = index_helper.access_id_in_token;
+  int token_stride = index_helper.token_stride;
+  int access_id = index_helper.access_id;
+  int access_stride = index_helper.access_stride;
+  int tot_access = index_helper.tot_access;
+  int tot_tokens = params.size_q / params.hidden_dim;
+  float4 clear_vec = get_neg_zero();
+
+  LamportComm<NRanks> comm(params.workspace, params.rank);
+  int clear_access = comm.clear_size / kElemsPerAccess<DType>;
+#if (defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 900))
+  asm volatile("griddepcontrol.wait;");
+#endif
+  for (int idx = access_id; idx < tot_access;
+       idx += access_stride, token_id += token_stride) {
+    alignas(16) DType vals[kElemsPerAccess<DType>];
+    float sum_variance = 0.F;
+    *reinterpret_cast<float4*>(vals) =
+        reinterpret_cast<float4*>(params.allreduce_in)[idx];
+#pragma unroll
+    for (int i = 0; i < kElemsPerAccess<DType>; ++i) {
+      sum_variance += static_cast<float>(vals[i]) * static_cast<float>(vals[i]);
+    }
+    blockReduceSumV2<float, 1>(&sum_variance);
+    if (is_neg_zero(sum_variance)) {
+      sum_variance = 0.F;
+    }
+    if (threadIdx.x == 0) {
+      for (int r = 0; r < NRanks; ++r) {
+        reinterpret_cast<float*>(
+            comm.data_bufs[r])[(params.rank * tot_tokens) + token_id] =
+            (sum_variance);
+      }
+    }
+
+    bool done = false;
+    float vars_all_ranks[NRanks];
+    while (!done) {
+      done = true;
+#pragma unroll
+      for (int r = 0; r < NRanks; ++r) {
+        vars_all_ranks[r] = ld_global_volatile(&reinterpret_cast<float*>(
+            comm.data_bufs[params.rank])[(r * tot_tokens) + token_id]);
+        done &= !is_neg_zero(vars_all_ranks[r]);
+      }
+    }
+    sum_variance = 0.F;
+#pragma unroll
+    for (int r = 0; r < NRanks; ++r) {
+      sum_variance += vars_all_ranks[r];
+    }
+
+    DType norm_weight[kElemsPerAccess<DType>];
+    *reinterpret_cast<typename ElemsPerAccess<DType>::vec_type*>(norm_weight) =
+        reinterpret_cast<typename ElemsPerAccess<DType>::vec_type*>(
+            params.rms_gamma)[access_id_in_token];
+
+#pragma unroll
+    for (int i = 0; i < kElemsPerAccess<DType>; ++i) {
+      vals[i] = static_cast<DType>(
+          static_cast<float>(vals[i]) *
+          rsqrtf(
+              (sum_variance / static_cast<float>(params.hidden_dim) / NRanks) +
+              params.rms_eps) *
+          static_cast<float>(norm_weight[i]));
+    }
+
+    reinterpret_cast<float4*>(params.rms_norm_out)[idx] =
+        *reinterpret_cast<float4*>(vals);
+  }
+  for (int idx = access_id; idx < clear_access; idx += access_stride) {
+    reinterpret_cast<float4*>(comm.clear_buf)[idx] = clear_vec;
+  }
+  comm.update(params.size_q * NRanks);
+#if (defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 900))
+  asm volatile("griddepcontrol.launch_dependents;");
+#endif
+}
+
+/**
+ * Float4 variant: process 4 rows at once, allreduce variance sums as float4 for
+ * better memory coalescing. sum_variance is always float; applies to all DTypes
+ * (half, bf16, float). When tot_tokens % 4 != 0, the last group pads rows with
+ * zeros; padded rows are not written to rms_norm_out. IsQK: when true, process
+ * Q+K in one loop with doubled comm buffer; when false, single-matrix (Q only).
+ */
+template <typename DType, int NRanks, int OriginQDim, int OriginKDim>
+__global__ void __launch_bounds__(1024)
+    minimax_reduce_qk_rms_kernel_lamport_float4(MiniMaxReduceRMSParams params) {
+  // Compile-time per-rank dimensions
+  constexpr int RankQDim = OriginQDim / NRanks;
+  constexpr int RankKDim = OriginKDim / NRanks;
+  // Threads needed to cover one row of Q / K with float4 accesses
+  constexpr int ThreadsPerRowQ = RankQDim / kElemsPerAccess<DType>;
+  constexpr int ThreadsPerRowK = RankKDim / kElemsPerAccess<DType>;
+  // Number of warps dedicated to Q / K
+  constexpr int NumWarpQ = (ThreadsPerRowQ + MINIMAX_REDUCE_RMS_WARP_SIZE - 1) /
+                           MINIMAX_REDUCE_RMS_WARP_SIZE;
+  constexpr int NumWarpK = (ThreadsPerRowK + MINIMAX_REDUCE_RMS_WARP_SIZE - 1) /
+                           MINIMAX_REDUCE_RMS_WARP_SIZE;
+
+  int tot_tokens = params.size_q / RankQDim;
+  int tot_groups = (tot_tokens + 3) / 4;  // ceiling; last group may be partial
+
+  // Memory strides for strided qkv tensors (elements -> float4-access units)
+  int access_stride_q = (params.stride_q > 0 ? params.stride_q : RankQDim) /
+                        kElemsPerAccess<DType>;
+  int access_stride_k = (params.stride_k > 0 ? params.stride_k : RankKDim) /
+                        kElemsPerAccess<DType>;
+  // Output strides: default to contiguous (hidden_dim / hidden_dim_k)
+  int access_stride_q_out =
+      (params.stride_q_out > 0 ? params.stride_q_out : params.hidden_dim) /
+      kElemsPerAccess<DType>;
+  int access_stride_k_out =
+      (params.stride_k_out > 0 ? params.stride_k_out : params.hidden_dim_k) /
+      kElemsPerAccess<DType>;
+
+#if (defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 900))
+  namespace cg = cooperative_groups;
+  cg::cluster_group cluster = cg::this_cluster();
+  cg::grid_group grid = cg::this_grid();
+  int group_id = grid.cluster_rank();
+  int access_id_in_token = cluster.thread_rank();
+  int group_stride = grid.num_clusters();
+#else
+  int group_id = blockIdx.x;
+  int access_id_in_token = threadIdx.x;
+  int group_stride = gridDim.x;
+#endif
+
+  bool is_q = (access_id_in_token < NumWarpQ * MINIMAX_REDUCE_RMS_WARP_SIZE);
+  int k_thread_idx =
+      access_id_in_token - (NumWarpQ * MINIMAX_REDUCE_RMS_WARP_SIZE);
+  bool is_valid_q = (access_id_in_token < ThreadsPerRowQ);
+  bool is_valid_k = (k_thread_idx >= 0 && k_thread_idx < ThreadsPerRowK);
+  float4 clear_vec = get_neg_zero();
+
+  // Shared memory for two-level block reduction and scale broadcast
+  __shared__ float block_reduce_sum[4][MINIMAX_REDUCE_RMS_WARP_SIZE + 1];
+  __shared__ float global_scale_q[4];
+  __shared__ float global_scale_k[4];
+
+  LamportComm<NRanks> comm(params.workspace, params.rank);
+
+  DType norm_weight[kElemsPerAccess<DType>]{};
+#if (defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 900))
+  asm volatile("griddepcontrol.wait;");
+#endif
+  if (is_q) {
+    if (is_valid_q) {
+      *reinterpret_cast<typename ElemsPerAccess<DType>::vec_type*>(
+          norm_weight) =
+          reinterpret_cast<typename ElemsPerAccess<DType>::vec_type const*>(
+              params.rms_gamma)[access_id_in_token];
+    }
+  } else {
+    if (is_valid_k) {
+      *reinterpret_cast<typename ElemsPerAccess<DType>::vec_type*>(
+          norm_weight) =
+          reinterpret_cast<typename ElemsPerAccess<DType>::vec_type const*>(
+              params.rms_gamma_k)[k_thread_idx];
+    }
+  }
+
+  // Main loop: process one group of 4 tokens per iteration.
+  for (int g = group_id; g < tot_groups; g += group_stride) {
+    alignas(16) DType vals[4][kElemsPerAccess<DType>]{};
+    float warp_sum_variance[4]{0.F, 0.F, 0.F, 0.F};
+
+    if (is_q) {
+#pragma unroll
+      for (int row = 0; row < 4; ++row) {
+        int token_r = g * 4 + row;
+        if (token_r >= tot_tokens || !is_valid_q) {
+          continue;
+        }
+        int idx_r = token_r * access_stride_q + access_id_in_token;
+        *reinterpret_cast<float4*>(&vals[row][0]) =
+            reinterpret_cast<float4 const*>(params.allreduce_in)[idx_r];
+#pragma unroll
+        for (int i = 0; i < kElemsPerAccess<DType>; ++i) {
+          float x = static_cast<float>(vals[row][i]);
+          warp_sum_variance[row] += x * x;
+        }
+      }
+    } else {
+#pragma unroll
+      for (int row = 0; row < 4; ++row) {
+        int token_r = g * 4 + row;
+        if (token_r >= tot_tokens || !is_valid_k) {
+          continue;
+        }
+        int idx_r = token_r * access_stride_k + k_thread_idx;
+        *reinterpret_cast<float4*>(&vals[row][0]) =
+            reinterpret_cast<float4 const*>(params.allreduce_in_k)[idx_r];
+#pragma unroll
+        for (int i = 0; i < kElemsPerAccess<DType>; ++i) {
+          float x = static_cast<float>(vals[row][i]);
+          warp_sum_variance[row] += x * x;
+        }
+      }
+    }
+
+    local_warp_reduce_sum_array<MINIMAX_REDUCE_RMS_WARP_SIZE, float, 4>(
+        warp_sum_variance);
+    // Warp lane 0 writes its warp's partial sum to shared memory
+    int lane = threadIdx.x & (MINIMAX_REDUCE_RMS_WARP_SIZE - 1);
+    if (lane == 0) {
+#pragma unroll
+      for (int t = 0; t < 4; ++t) {
+        block_reduce_sum[t][threadIdx.x / MINIMAX_REDUCE_RMS_WARP_SIZE] =
+            warp_sum_variance[t];
+      }
+    }
+    __syncthreads();
+
+    int tid = threadIdx.x;
+
+    if (tid < MINIMAX_REDUCE_RMS_WARP_SIZE) {
+      constexpr int kNumWarpQPow2 =
+          (next_pow2(NumWarpQ) > NRanks) ? next_pow2(NumWarpQ) : NRanks;
+      float local_sum[4];
+#pragma unroll
+      for (int t = 0; t < 4; ++t) {
+        local_sum[t] = (tid < NumWarpQ) ? block_reduce_sum[t][tid] : 0.F;
+      }
+      // After this, all kNumWarpQPow2 lanes (including tid 0..NRanks-1) have
+      // the total Q sum-of-squares for all 4 tokens.
+      local_warp_reduce_sum_array<kNumWarpQPow2, float, 4>(local_sum);
+
+      if (tid < NRanks) {
+#pragma unroll
+        for (int t = 0; t < 4; ++t) {
+          if (is_neg_zero(local_sum[t])) {
+            local_sum[t] = 0.F;
+          }
+        }
+        // Parallel push: thread tid writes this rank's Q sum to rank tid's buf
+        reinterpret_cast<float4*>(
+            comm.data_bufs[tid])[(params.rank * tot_groups * 2) + (2 * g)] =
+            *reinterpret_cast<float4*>(local_sum);
+
+        // Parallel pull: thread tid reads rank tid's contribution from
+        // this rank's (params.rank's) buffer
+        bool done = false;
+        float4 var_all_ranks;
+        while (!done) {
+          done = true;
+          var_all_ranks = ld_global_volatile(&reinterpret_cast<float4*>(
+              comm.data_bufs[params.rank])[(tid * tot_groups * 2) + (2 * g)]);
+          done &= !is_neg_zero(var_all_ranks);
+        }
+
+        // Warp-level allreduce: each of the NRanks threads holds one rank's
+        // partial sum; after this all NRanks threads have the global total.
+        constexpr uint32_t kQActiveMask = (1u << NRanks) - 1u;
+        local_warp_reduce_sum_array<NRanks, float, 4>(
+            reinterpret_cast<float*>(&var_all_ranks), kQActiveMask);
+
+        // Thread 0 computes rsqrt with compile-time Dim and writes to smem
+        if (tid == 0) {
+          *reinterpret_cast<float4*>(global_scale_q) =
+              rms_rsqrt<OriginQDim>(var_all_ranks, params.rms_eps);
+        }
+      }
+    } else if (tid >= MINIMAX_REDUCE_RMS_WARP_SIZE * NumWarpQ &&
+               tid < MINIMAX_REDUCE_RMS_WARP_SIZE * (NumWarpQ + 1)) {
+      // --- K leader warp ---
+      constexpr int kNumWarpKPow2 =
+          (next_pow2(NumWarpK) > NRanks) ? next_pow2(NumWarpK) : NRanks;
+      float local_sum[4];
+#pragma unroll
+      for (int t = 0; t < 4; ++t) {
+        local_sum[t] = (k_thread_idx < NumWarpK)
+                           ? block_reduce_sum[t][NumWarpQ + k_thread_idx]
+                           : 0.F;
+      }
+      local_warp_reduce_sum_array<kNumWarpKPow2, float, 4>(local_sum);
+
+      if (k_thread_idx < NRanks) {
+#pragma unroll
+        for (int t = 0; t < 4; ++t) {
+          if (is_neg_zero(local_sum[t])) {
+            local_sum[t] = 0.F;
+          }
+        }
+        reinterpret_cast<float4*>(
+            comm.data_bufs[k_thread_idx])[(params.rank * tot_groups * 2) +
+                                          (2 * g + 1)] =
+            *reinterpret_cast<float4*>(local_sum);
+
+        bool done = false;
+        float4 var_all_ranks;
+        while (!done) {
+          done = true;
+          var_all_ranks = ld_global_volatile(&reinterpret_cast<float4*>(
+              comm.data_bufs[params.rank])[(k_thread_idx * tot_groups * 2) +
+                                           (2 * g + 1)]);
+          done &= !is_neg_zero(var_all_ranks);
+        }
+
+        constexpr uint32_t kKActiveMask = (1u << NRanks) - 1u;
+        local_warp_reduce_sum_array<NRanks, float, 4>(
+            reinterpret_cast<float*>(&var_all_ranks), kKActiveMask);
+
+        if (k_thread_idx == 0) {
+          *reinterpret_cast<float4*>(global_scale_k) =
+              rms_rsqrt<OriginKDim>(var_all_ranks, params.rms_eps);
+        }
+      }
+    }
+    __syncthreads();
+
+    if (is_q) {
+#pragma unroll
+      for (int t = 0; t < 4; ++t) {
+        warp_sum_variance[t] = global_scale_q[t];
+      }
+#pragma unroll
+      for (int r = 0; r < 4; ++r) {
+#pragma unroll
+        for (int i = 0; i < kElemsPerAccess<DType>; ++i) {
+          vals[r][i] = static_cast<DType>(static_cast<float>(vals[r][i]) *
+                                          warp_sum_variance[r] *
+                                          static_cast<float>(norm_weight[i]));
+        }
+        int token_r = g * 4 + r;
+        if (token_r >= tot_tokens || !is_valid_q) {
+          continue;
+        }
+        int idx_out = token_r * access_stride_q_out + access_id_in_token;
+        reinterpret_cast<float4*>(params.rms_norm_out)[idx_out] =
+            *reinterpret_cast<float4*>(&vals[r][0]);
+      }
+    } else {
+#pragma unroll
+      for (int t = 0; t < 4; ++t) {
+        warp_sum_variance[t] = global_scale_k[t];
+      }
+#pragma unroll
+      for (int r = 0; r < 4; ++r) {
+#pragma unroll
+        for (int i = 0; i < kElemsPerAccess<DType>; ++i) {
+          vals[r][i] = static_cast<DType>(static_cast<float>(vals[r][i]) *
+                                          warp_sum_variance[r] *
+                                          static_cast<float>(norm_weight[i]));
+        }
+        int token_r = g * 4 + r;
+        if (token_r >= tot_tokens || !is_valid_k) {
+          continue;
+        }
+        int idx_out = token_r * access_stride_k_out + k_thread_idx;
+        reinterpret_cast<float4*>(params.rms_norm_out_k)[idx_out] =
+            *reinterpret_cast<float4*>(&vals[r][0]);
+      }
+    }
+  }  // end group loop
+#if (defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 900))
+  asm volatile("griddepcontrol.launch_dependents;");
+#endif
+
+  int clear_access = static_cast<int>(comm.clear_size / kElemsPerAccess<DType>);
+  int clear_stride = group_stride * blockDim.x;
+  for (int idx = group_id * blockDim.x + threadIdx.x; idx < clear_access;
+       idx += clear_stride) {
+    reinterpret_cast<float4*>(comm.clear_buf)[idx] = clear_vec;
+  }
+
+  comm.update(static_cast<int64_t>(2) * tot_groups * kElemsPerAccess<DType> *
+              NRanks);
+}
+
+int get_sm_count() {
+  static int sm_count = 0;
+  if (sm_count == 0) {
+    int device_id;
+    CUDA_CHECK(cudaGetDevice(&device_id));
+    cudaDeviceProp device_prop;
+    cudaGetDeviceProperties(&device_prop, device_id);
+    sm_count = device_prop.multiProcessorCount;
+  }
+  return sm_count;
+}
+
+inline int getSMVersion(bool queryRealSmArch = false) {
+  int device{-1};
+  CUDA_CHECK(cudaGetDevice(&device));
+  int sm_major = 0;
+  int sm_minor = 0;
+  CUDA_CHECK(cudaDeviceGetAttribute(&sm_major,
+                                    cudaDevAttrComputeCapabilityMajor, device));
+  CUDA_CHECK(cudaDeviceGetAttribute(&sm_minor,
+                                    cudaDevAttrComputeCapabilityMinor, device));
+  int sm = sm_major * 10 + sm_minor;
+  if (sm == 121 && !queryRealSmArch) {
+    return 120;
+  }
+  return sm;
+}
+
+template <typename KernelFunc>
+int get_max_active_blocks(KernelFunc kernel, int block_size,
+                          int dynamic_smem = 0) {
+  int max_active = 0;
+  CUDA_CHECK(cudaOccupancyMaxActiveBlocksPerMultiprocessor(
+      &max_active, kernel, block_size, dynamic_smem));
+  return std::max(max_active, 1);
+}
+
+template <typename DType, int NRanks>
+void minimax_reduce_rms_kernel_launcher(MiniMaxReduceRMSParams const& params) {
+  static int SM = getSMVersion();
+  int token_num = params.size_q / params.hidden_dim;
+  int sm_count = get_sm_count();
+  int cluster_size = 1;
+  int cluster_num = token_num;
+  int threads_per_token = params.hidden_dim / kElemsPerAccess<DType>;
+  int block_size = threads_per_token;
+
+  int max_blocks_per_sm = get_max_active_blocks(
+      minimax_reduce_rms_kernel_lamport<DType, NRanks>, block_size);
+  int max_grid = max_blocks_per_sm * sm_count;
+
+  int grid_size =
+      (std::min(max_grid, cluster_num * cluster_size) / cluster_size) *
+      cluster_size;
+
+  cudaLaunchConfig_t cfg;
+  cfg.gridDim = grid_size;
+  cfg.blockDim = block_size;
+  cfg.dynamicSmemBytes = 0;
+  cfg.stream = params.stream;
+
+  cudaLaunchAttribute attribute[2];
+  attribute[0].id = cudaLaunchAttributeProgrammaticStreamSerialization;
+  attribute[0].val.programmaticStreamSerializationAllowed = 1;
+  attribute[1].id = cudaLaunchAttributeClusterDimension;
+  attribute[1].val.clusterDim.x = cluster_size;
+  attribute[1].val.clusterDim.y = 1;
+  attribute[1].val.clusterDim.z = 1;
+  cfg.attrs = attribute;
+  cfg.numAttrs = SM >= 90 ? 2 : 0;
+
+  CUDA_CHECK(cudaLaunchKernelEx(
+      &cfg, minimax_reduce_rms_kernel_lamport<DType, NRanks>, params));
+}
+
+template <typename DType, int NRanks, int OriginQDim, int OriginKDim>
+void minimax_reduce_rms_kernel_launcher_float4(
+    MiniMaxReduceRMSParams const& params) {
+  TORCH_CHECK(params.size_q % params.hidden_dim == 0);
+  TORCH_CHECK(params.hidden_dim % kElemsPerAccess<DType> == 0);
+  if (params.stride_q > 0) {
+    TORCH_CHECK(params.stride_q % kElemsPerAccess<DType> == 0);
+  }
+  TORCH_CHECK(params.allreduce_in_k != nullptr,
+              "float4 QK kernel requires K input");
+  TORCH_CHECK(params.hidden_dim >= params.hidden_dim_k);
+  TORCH_CHECK(params.size_k % params.hidden_dim_k == 0);
+  TORCH_CHECK(params.hidden_dim_k % kElemsPerAccess<DType> == 0);
+  TORCH_CHECK(params.size_q / params.hidden_dim ==
+              params.size_k / params.hidden_dim_k);
+  if (params.stride_k > 0) {
+    TORCH_CHECK(params.stride_k % kElemsPerAccess<DType> == 0);
+  }
+
+  int token_num = params.size_q / params.hidden_dim;
+  int tot_groups = (token_num + 3) / 4;
+  if (tot_groups == 0) {
+    return;
+  }
+
+  static int SM = getSMVersion();
+  int sm_count = get_sm_count();
+  int cluster_size = 1;
+  int cluster_num = tot_groups;
+
+  int access_per_row_q = params.hidden_dim / kElemsPerAccess<DType>;
+  int access_per_row_k = params.hidden_dim_k / kElemsPerAccess<DType>;
+
+  // Round each section up to a warp boundary
+  auto divUp = [](int a, int b) { return (a + b - 1) / b * b; };
+  int block_size = divUp(access_per_row_q, MINIMAX_REDUCE_RMS_WARP_SIZE) +
+                   divUp(access_per_row_k, MINIMAX_REDUCE_RMS_WARP_SIZE);
+
+  auto kfn =
+      minimax_reduce_qk_rms_kernel_lamport_float4<DType, NRanks, OriginQDim,
+                                                  OriginKDim>;
+
+  int max_blocks_per_sm = get_max_active_blocks(kfn, block_size);
+  int max_grid = max_blocks_per_sm * sm_count;
+  int grid_size =
+      (std::min(max_grid, cluster_num * cluster_size) / cluster_size) *
+      cluster_size;
+
+  cudaLaunchConfig_t cfg;
+  cfg.gridDim = grid_size;
+  cfg.blockDim = block_size;
+  cfg.dynamicSmemBytes = 0;
+  cfg.stream = params.stream;
+
+  cudaLaunchAttribute attribute[2];
+  attribute[0].id = cudaLaunchAttributeProgrammaticStreamSerialization;
+  attribute[0].val.programmaticStreamSerializationAllowed = 1;
+  attribute[1].id = cudaLaunchAttributeClusterDimension;
+  attribute[1].val.clusterDim.x = cluster_size;
+  attribute[1].val.clusterDim.y = 1;
+  attribute[1].val.clusterDim.z = 1;
+  cfg.attrs = attribute;
+  cfg.numAttrs = SM >= 90 ? 2 : 0;
+
+  CUDA_CHECK(cudaLaunchKernelEx(&cfg, kfn, params));
+}
+
+template <int NRanks>
+void dispatch_dtype(MiniMaxReduceRMSParams const& params) {
+  // Use the optimized QK float4 kernel when:
+  //  - K input is present, AND
+  //  - the full (NRanks * per-rank) dimensions match the MiniMax M2 shape.
+  // Otherwise fall back to the scalar kernel.
+  bool use_float4 = (params.allreduce_in_k != nullptr) &&
+                    (params.hidden_dim * params.nranks == 6144) &&
+                    (params.hidden_dim_k * params.nranks == 1024);
+
+  if (params.dtype == at::ScalarType::Half) {
+    if (use_float4) {
+      minimax_reduce_rms_kernel_launcher_float4<half, NRanks, 6144, 1024>(
+          params);
+    } else {
+      minimax_reduce_rms_kernel_launcher<half, NRanks>(params);
+    }
+  } else if (params.dtype == at::ScalarType::BFloat16) {
+    if (use_float4) {
+      minimax_reduce_rms_kernel_launcher_float4<__nv_bfloat16, NRanks, 6144,
+                                                1024>(params);
+    } else {
+      minimax_reduce_rms_kernel_launcher<__nv_bfloat16, NRanks>(params);
+    }
+  } else if (params.dtype == at::ScalarType::Float) {
+    if (use_float4) {
+      minimax_reduce_rms_kernel_launcher_float4<float, NRanks, 6144, 1024>(
+          params);
+    } else {
+      minimax_reduce_rms_kernel_launcher<float, NRanks>(params);
+    }
+  } else {
+    TORCH_CHECK(false, "Unsupported data type for minimax_reduce_rms_op");
+  }
+}
+
+void minimax_reduce_rms_op(MiniMaxReduceRMSParams const& params) {
+  if (params.nranks == 2) {
+    dispatch_dtype<2>(params);
+  } else if (params.nranks == 4) {
+    dispatch_dtype<4>(params);
+  } else if (params.nranks == 8) {
+    dispatch_dtype<8>(params);
+  } else if (params.nranks == 16) {
+    dispatch_dtype<16>(params);
+  } else {
+    TORCH_CHECK(false, "minimax_reduce_rms_op: unsupported ranks number!");
+  }
+}
+}  // namespace tensorrt_llm
+}  // namespace vllm
+
+torch::Tensor minimax_allreduce_rms(torch::Tensor const& input,
+                                    torch::Tensor const& norm_weight,
+                                    torch::Tensor workspace, int64_t const rank,
+                                    int64_t const nranks, double const eps) {
+  auto allreduce_params = vllm::tensorrt_llm::MiniMaxReduceRMSParams();
+
+  allreduce_params.nranks = static_cast<int>(nranks);
+  allreduce_params.rank = static_cast<int>(rank);
+  allreduce_params.dtype = input.scalar_type();
+  allreduce_params.size_q = static_cast<int>(input.numel());
+  allreduce_params.hidden_dim = static_cast<int>(input.size(-1));
+  allreduce_params.stride_q = allreduce_params.hidden_dim;
+  allreduce_params.workspace =
+      reinterpret_cast<void**>(workspace.mutable_data_ptr());
+  allreduce_params.allreduce_in = input.data_ptr();
+  allreduce_params.rms_gamma = norm_weight.data_ptr();
+  allreduce_params.rms_eps = static_cast<float>(eps);
+  allreduce_params.stream = at::cuda::getCurrentCUDAStream(input.get_device());
+
+  torch::Tensor rms_norm_out = torch::empty_like(input);
+  allreduce_params.rms_norm_out = rms_norm_out.mutable_data_ptr();
+
+  vllm::tensorrt_llm::minimax_reduce_rms_op(allreduce_params);
+
+  return rms_norm_out;
+}
+
+std::tuple<torch::Tensor, torch::Tensor> minimax_allreduce_rms_qk(
+    torch::Tensor qkv, torch::Tensor const& norm_weight_q,
+    torch::Tensor const& norm_weight_k, torch::Tensor workspace,
+    int64_t const q_size, int64_t const kv_size, int64_t const rank,
+    int64_t const nranks, double const eps) {
+  TORCH_CHECK(qkv.dim() == 2, "minimax_allreduce_rms_qk: qkv must be 2D");
+  TORCH_CHECK(qkv.is_contiguous(),
+              "minimax_allreduce_rms_qk: qkv must be contiguous");
+  int64_t qkv_dim = qkv.size(-1);
+  TORCH_CHECK(qkv_dim == q_size + 2 * kv_size,
+              "minimax_allreduce_rms_qk: qkv last dim must equal "
+              "q_size + 2 * kv_size");
+  TORCH_CHECK(rank < nranks,
+              "minimax_allreduce_rms_qk: rank must be less than nranks");
+
+  int64_t num_tokens = qkv.size(0);
+  int elem_bytes = qkv.element_size();
+
+  torch::Tensor q_out = torch::empty({num_tokens, q_size}, qkv.options());
+  torch::Tensor k_out = torch::empty({num_tokens, kv_size}, qkv.options());
+
+  auto params = vllm::tensorrt_llm::MiniMaxReduceRMSParams();
+  params.nranks = static_cast<int>(nranks);
+  params.rank = static_cast<int>(rank);
+  params.dtype = qkv.scalar_type();
+  params.size_q = static_cast<int>(num_tokens * q_size);
+  params.hidden_dim = static_cast<int>(q_size);
+  params.size_k = static_cast<int>(num_tokens * kv_size);
+  params.hidden_dim_k = static_cast<int>(kv_size);
+  params.stride_q = static_cast<int>(qkv_dim);
+  params.stride_k = static_cast<int>(qkv_dim);
+  params.stride_q_out = 0;  // q_out is contiguous; kernel uses hidden_dim
+  params.stride_k_out = 0;  // k_out is contiguous; kernel uses hidden_dim_k
+  params.workspace = reinterpret_cast<void**>(workspace.mutable_data_ptr());
+
+  uint8_t* base = static_cast<uint8_t*>(qkv.data_ptr());
+  params.allreduce_in = base;
+  params.allreduce_in_k = base + q_size * elem_bytes;
+  params.rms_gamma = norm_weight_q.data_ptr();
+  params.rms_gamma_k = norm_weight_k.data_ptr();
+  params.rms_eps = static_cast<float>(eps);
+  params.stream = at::cuda::getCurrentCUDAStream(qkv.get_device());
+
+  params.rms_norm_out = q_out.mutable_data_ptr();
+  params.rms_norm_out_k = k_out.mutable_data_ptr();
+
+  vllm::tensorrt_llm::minimax_reduce_rms_op(params);
+  return {q_out, k_out};
+}

csrc/minimax_reduce_rms_kernel.h

+/*
+ * Copyright (c) 2026, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#pragma once
+
+#include <cuda_bf16.h>
+#include <cuda_fp16.h>
+
+#include <torch/types.h>
+
+namespace vllm {
+namespace tensorrt_llm {
+
+template <typename DType>
+struct ElemsPerAccess;
+
+template <>
+struct ElemsPerAccess<half> {
+  static constexpr int value = 8;
+  using vec_type = float4;
+};
+
+template <>
+struct ElemsPerAccess<nv_bfloat16> {
+  static constexpr int value = 8;
+  using vec_type = float4;
+};
+
+template <>
+struct ElemsPerAccess<float> {
+  static constexpr int value = 4;
+  using vec_type = float4;
+};
+
+template <typename DType>
+static constexpr int kElemsPerAccess = ElemsPerAccess<DType>::value;
+
+struct MiniMaxReduceRMSParams {
+  int nranks{};
+  int rank{};
+  at::ScalarType dtype{at::ScalarType::Undefined};
+  int size_q{};
+  int hidden_dim{};
+  int size_k{};
+  int hidden_dim_k{};
+  int stride_q{};  // row stride for q input (elements); when > hidden_dim,
+                   // q is part of a wider qkv tensor
+  int stride_k{};  // row stride for k input (elements); when > hidden_dim_k,
+                   // k is part of a wider qkv tensor
+  int stride_q_out{};  // row stride for q output (elements); 0 = contiguous
+  int stride_k_out{};  // row stride for k output (elements); 0 = contiguous
+  void** workspace{};
+  void* allreduce_in{};
+  void* rms_norm_out{};
+  void* rms_gamma{};
+  void* allreduce_in_k{};
+  void* rms_norm_out_k{};
+  void* rms_gamma_k{};
+  float rms_eps{};
+  cudaStream_t stream{};
+};
+
+void minimax_reduce_rms_op(MiniMaxReduceRMSParams const& params);
+
+}  // namespace tensorrt_llm
+}  // namespace vllm

csrc/ops.h

@@ -309,3 +309,15 @@ int64_t qr_max_size();
 void dsv3_fused_a_gemm(torch::Tensor& output, torch::Tensor const& mat_a,
                        torch::Tensor const& mat_b);
 #endif
+
+#ifndef USE_ROCM
+torch::Tensor minimax_allreduce_rms(torch::Tensor const& input,
+                                    torch::Tensor const& norm_weight,
+                                    torch::Tensor workspace, int64_t const rank,
+                                    int64_t const nranks, double const eps);
+std::tuple<torch::Tensor, torch::Tensor> minimax_allreduce_rms_qk(
+    torch::Tensor qkv, torch::Tensor const& norm_weight_q,
+    torch::Tensor const& norm_weight_k, torch::Tensor workspace,
+    int64_t const q_size, int64_t const kv_size, int64_t const rank,
+    int64_t const nranks, double const eps);
+#endif
\ No newline at end of file

csrc/torch_bindings.cpp

@@ -496,6 +496,29 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
       "Tensor? b_qzeros, "
       "SymInt n, SymInt group_size, SymInt sm_count, SymInt sm_version, SymInt "
       "CUBLAS_M_THRESHOLD, bool has_zp, bool n32k16_reorder) -> Tensor");
+
+  ops.def(
+      "minimax_allreduce_rms("
+      "Tensor input,"
+      "Tensor norm_weight,"
+      "Tensor workspace,"
+      "int rank,"
+      "int nranks,"
+      "float eps) -> Tensor");
+  ops.impl("minimax_allreduce_rms", torch::kCUDA, &minimax_allreduce_rms);
+  ops.def(
+      "minimax_allreduce_rms_qk("
+      "Tensor qkv,"
+      "Tensor norm_weight_q,"
+      "Tensor norm_weight_k,"
+      "Tensor workspace,"
+      "int q_size,"
+      "int kv_size,"
+      "int rank,"
+      "int nranks,"
+      "float eps) -> (Tensor, Tensor)");
+  ops.impl("minimax_allreduce_rms_qk", torch::kCUDA, &minimax_allreduce_rms_qk);
+
   //  conditionally compiled so impl in source file
 #endif
 }

tests/kernels/core/test_minimax_reduce_rms.py

+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+"""Tests for MiniMax QK RMS-norm: NCCL reference vs Lamport fused kernel."""
+
+import pytest
+import torch
+import torch.nn as nn
+from torch.multiprocessing import spawn
+
+from tests.kernels.utils import opcheck
+from tests.utils import ensure_current_vllm_config, init_test_distributed_environment
+from vllm.distributed import cleanup_dist_env_and_memory
+from vllm.model_executor.layers.mamba.linear_attn import MiniMaxText01RMSNormTP
+from vllm.platforms import current_platform
+from vllm.utils.network_utils import get_open_port
+from vllm.utils.torch_utils import set_random_seed
+
+
+@ensure_current_vllm_config()
+def _worker_forward_qk(
+    local_rank,
+    world_size,
+    port,
+    num_tokens,
+    hidden_q_full,
+    hidden_k_full,
+    dtype,
+    seed,
+    eps,
+):
+    """Per-rank worker: compare NCCL allreduce path vs Lamport fused kernel."""
+
+    if not hasattr(torch.ops._C, "minimax_allreduce_rms_qk"):
+        cleanup_dist_env_and_memory()
+        return
+    device = torch.device(f"cuda:{local_rank}")
+    torch.accelerator.set_device_index(device)
+    init_test_distributed_environment(
+        world_size, 1, local_rank, port, local_rank=local_rank
+    )
+
+    hq = hidden_q_full // world_size
+    hk = hidden_k_full // world_size
+
+    q_norm = MiniMaxText01RMSNormTP(hidden_q_full, eps=eps).cuda()
+    k_norm = MiniMaxText01RMSNormTP(hidden_k_full, eps=eps).cuda()
+
+    set_random_seed(seed)
+    qw = torch.randn(hidden_q_full, dtype=dtype, device="cuda")
+    kw = torch.randn(hidden_k_full, dtype=dtype, device="cuda")
+    q_norm.weight = nn.Parameter(qw[local_rank * hq : (local_rank + 1) * hq])
+    k_norm.weight = nn.Parameter(kw[local_rank * hk : (local_rank + 1) * hk])
+
+    torch.manual_seed(seed + 1000 + local_rank)
+    qkv = torch.randn(num_tokens, hq + hk + hk, dtype=dtype, device="cuda")
+
+    q_ref, k_ref, v_ref = qkv.clone().split([hq, hk, hk], dim=-1)
+    ref_q, ref_k = MiniMaxText01RMSNormTP.forward_qk(q_norm, k_norm, q_ref, k_ref)
+
+    # Set up Lamport workspace.
+    from vllm.distributed.parallel_state import get_tp_group
+    from vllm.model_executor.layers.mamba.lamport_workspace import (
+        get_allreduce_workspace,
+    )
+
+    workspace = get_allreduce_workspace(
+        rank=local_rank,
+        world_size=world_size,
+        max_tokens=num_tokens,
+        process_group=get_tp_group().cpu_group,
+    )
+
+    opcheck(
+        torch.ops._C.minimax_allreduce_rms_qk,
+        (
+            qkv.clone(),
+            q_norm.weight,
+            k_norm.weight,
+            workspace,
+            hq,
+            hk,
+            local_rank,
+            world_size,
+            eps,
+        ),
+    )
+    fused_q, fused_k = torch.ops._C.minimax_allreduce_rms_qk(
+        qkv.clone(),
+        q_norm.weight,
+        k_norm.weight,
+        workspace,
+        hq,
+        hk,
+        local_rank,
+        world_size,
+        eps,
+    )
+    _, _, fused_v = qkv.split([hq, hk, hk], dim=-1)
+    torch.accelerator.synchronize()
+
+    torch.testing.assert_close(
+        fused_q,
+        ref_q,
+        atol=3e-2,
+        rtol=3e-2,
+    )
+    torch.testing.assert_close(fused_k, ref_k, atol=3e-2, rtol=3e-2)
+
+    cleanup_dist_env_and_memory()
+
+
+@pytest.mark.skipif(
+    not current_platform.is_cuda(),
+    reason="CUDA required",
+)
+@pytest.mark.parametrize("world_size", [2, 4, 8])
+@pytest.mark.parametrize("num_tokens", [1, 128, 333])
+@pytest.mark.parametrize(
+    "hidden_dims",
+    [(6144, 1024)],
+)
+@pytest.mark.parametrize("dtype", [torch.bfloat16, torch.float16])
+@pytest.mark.parametrize("eps", [1e-6])
+@pytest.mark.parametrize("seed", [42])
+def test_minimax_reduce_rms_qk(
+    world_size,
+    num_tokens,
+    hidden_dims,
+    dtype,
+    eps,
+    seed,
+):
+    num_gpus = current_platform.device_count()
+    if num_gpus < world_size:
+        pytest.skip(f"Need >= {world_size} GPUs, have {num_gpus}")
+    hidden_q_full, hidden_k_full = hidden_dims
+    port = str(get_open_port())
+    spawn(
+        _worker_forward_qk,
+        args=(
+            world_size,
+            port,
+            num_tokens,
+            hidden_q_full,
+            hidden_k_full,
+            dtype,
+            seed,
+            eps,
+        ),
+        nprocs=world_size,
+        join=True,
+    )

vllm/_custom_ops.py

@@ -3491,3 +3491,38 @@ if hasattr(torch.ops._C, "hadacore_transform"):
     @register_fake("_C::hadacore_transform")
     def _hadacore_transform_fake(x: torch.Tensor, inplace: bool) -> torch.Tensor:
         return torch.empty_like(x) if not inplace else x
+
+
+if hasattr(torch.ops._C, "minimax_allreduce_rms"):
+
+    @register_fake("_C::minimax_allreduce_rms")
+    def _minimax_allreduce_rms_fake(
+        input: torch.Tensor,
+        norm_weight: torch.Tensor,
+        workspace: torch.Tensor,
+        rank: int,
+        nranks: int,
+        eps: float,
+    ) -> torch.Tensor:
+        return torch.empty_like(input)
+
+
+if hasattr(torch.ops._C, "minimax_allreduce_rms_qk"):
+
+    @register_fake("_C::minimax_allreduce_rms_qk")
+    def _minimax_allreduce_rms_qk_fake(
+        qkv: torch.Tensor,
+        norm_weight_q: torch.Tensor,
+        norm_weight_k: torch.Tensor,
+        workspace: torch.Tensor,
+        q_size: int,
+        kv_size: int,
+        rank: int,
+        nranks: int,
+        eps: float,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        token_num = qkv.shape[0]
+        return (
+            torch.empty([token_num, q_size], dtype=qkv.dtype, device=qkv.device),
+            torch.empty([token_num, kv_size], dtype=qkv.dtype, device=qkv.device),
+        )

vllm/compilation/passes/fusion/minimax_qk_norm_fusion.py

+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+"""
+Fusion pass: replace MiniMax QK allreduce + RMS norm with the Lamport
+fused kernel (minimax_allreduce_rms_qk) for decode-size batches.
+
+Pattern (inlined forward_qk in compiled graph):
+    q, k, v = qkv.split([q_size, kv_size, kv_size], -1)
+    q_fp32 = q.to(float32); k_fp32 = k.to(float32)
+    q_var = q_fp32.pow(2).mean(-1, keepdim=True)
+    k_var = k_fp32.pow(2).mean(-1, keepdim=True)
+    qk_var = cat([q_var, k_var], -1)
+    qk_var = allreduce(qk_var) / tp_world
+    q_var, k_var = qk_var.chunk(2, -1)
+    q_out = (q_fp32 * rsqrt(q_var + eps) * q_weight).to(orig_dtype)
+    k_out = (k_fp32 * rsqrt(k_var + eps) * k_weight).to(orig_dtype)
+    return q_out, k_out, v
+
+Replacement (pure, no in-place on qkv/q/k):
+    q_out, k_out = minimax_qk_norm_fused(qkv, q_weight, k_weight, workspace, ...)
+    v = qkv.split([q_size, kv_size, kv_size], -1)[2]
+    return q_out, k_out, v
+
+is_applicable_for_range: only fires for compile_range.end <= max_decode_tokens
+so that large prefill batches fall through to the original forward_qk (= main).
+"""
+
+import torch
+import torch._inductor.pattern_matcher as pm
+import torch.fx as fx
+from torch._inductor.pattern_matcher import PatternMatcherPass
+
+from vllm.config import VllmConfig
+from vllm.config.utils import Range
+from vllm.distributed import tensor_model_parallel_all_reduce
+from vllm.distributed.parallel_state import (
+    get_tensor_model_parallel_rank,
+    get_tensor_model_parallel_world_size,
+)
+from vllm.logger import init_logger
+from vllm.utils.torch_utils import direct_register_custom_op
+
+from ..inductor_pass import enable_fake_mode
+from ..vllm_inductor_pass import VllmInductorPass, VllmPatternMatcherPass
+
+logger = init_logger(__name__)
+
+MAX_TOKEN_NUM = 2048
+
+_MINIMAX_QK_NORM_FUSED_OP = None
+if hasattr(torch.ops._C, "minimax_allreduce_rms_qk"):
+
+    def _minimax_qk_norm_fused(
+        qkv: torch.Tensor,
+        norm_weight_q: torch.Tensor,
+        norm_weight_k: torch.Tensor,
+        q_size: int,
+        kv_size: int,
+        rank: int,
+        nranks: int,
+        eps: float,
+        max_tokens: int,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        from vllm.distributed.parallel_state import get_tp_group
+        from vllm.model_executor.layers.mamba.lamport_workspace import (
+            get_allreduce_workspace,
+        )
+
+        workspace = get_allreduce_workspace(
+            rank=rank,
+            world_size=nranks,
+            max_tokens=max_tokens,
+            process_group=get_tp_group().cpu_group,
+        )
+        return torch.ops._C.minimax_allreduce_rms_qk(
+            qkv,
+            norm_weight_q,
+            norm_weight_k,
+            workspace,
+            q_size,
+            kv_size,
+            rank,
+            nranks,
+            eps,
+        )
+
+    def _minimax_qk_norm_fused_fake(
+        qkv: torch.Tensor,
+        norm_weight_q: torch.Tensor,
+        norm_weight_k: torch.Tensor,
+        q_size: int,
+        kv_size: int,
+        rank: int,
+        nranks: int,
+        eps: float,
+        max_tokens: int,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        T = qkv.shape[0]
+        return (
+            torch.empty([T, q_size], dtype=qkv.dtype, device=qkv.device),
+            torch.empty([T, kv_size], dtype=qkv.dtype, device=qkv.device),
+        )
+
+    direct_register_custom_op(
+        op_name="minimax_qk_norm_fused",
+        op_func=_minimax_qk_norm_fused,
+        fake_impl=_minimax_qk_norm_fused_fake,
+        mutates_args=[],
+    )
+    _MINIMAX_QK_NORM_FUSED_OP = torch.ops.vllm.minimax_qk_norm_fused.default
+
+
+class MiniMaxQKNormPattern:
+    """
+    Match the forward_qk allreduce+rms pattern and replace with Lamport kernel.
+    """
+
+    def __init__(
+        self,
+        q_size: int,
+        kv_size: int,
+        eps: float,
+        tp_world: int,
+        tp_rank: int,
+        max_tokens: int,
+        dtype: torch.dtype,
+        device: str | None,
+    ) -> None:
+        self.q_size = q_size
+        self.kv_size = kv_size
+        self.eps = eps
+        self.tp_world = tp_world
+        self.tp_rank = tp_rank
+        self.max_tokens = max_tokens
+        self.dtype = dtype
+        self.device = device
+
+    def get_inputs(self) -> list[torch.Tensor]:
+        T = 4
+        qkv = torch.empty(
+            [T, self.q_size + 2 * self.kv_size],
+            device=self.device,
+            dtype=self.dtype,
+        )
+        q_weight = torch.empty([self.q_size], device=self.device, dtype=self.dtype)
+        k_weight = torch.empty([self.kv_size], device=self.device, dtype=self.dtype)
+        return [qkv, q_weight, k_weight]
+
+    def register(self, pm_pass: PatternMatcherPass) -> None:
+        q_size = self.q_size
+        kv_size = self.kv_size
+        eps = self.eps
+        tp_world = self.tp_world
+        max_tokens = self.max_tokens
+        tp_rank = self.tp_rank
+        dtype = self.dtype
+
+        def pattern(
+            qkv: torch.Tensor,
+            q_weight: torch.Tensor,
+            k_weight: torch.Tensor,
+        ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+            q, k, v = qkv.split([q_size, kv_size, kv_size], dim=-1)
+            q_fp32 = q.to(torch.float32)
+            k_fp32 = k.to(torch.float32)
+            q_var = q_fp32.pow(2).mean(dim=-1, keepdim=True)
+            k_var = k_fp32.pow(2).mean(dim=-1, keepdim=True)
+            qk_var = torch.cat([q_var, k_var], dim=-1)
+            qk_var = tensor_model_parallel_all_reduce(qk_var) / tp_world
+            q_var, k_var = qk_var.chunk(2, dim=-1)
+            q_out = (q_fp32 * torch.rsqrt(q_var + eps) * q_weight).to(dtype)
+            k_out = (k_fp32 * torch.rsqrt(k_var + eps) * k_weight).to(dtype)
+            return q_out, k_out, v
+
+        def replacement(
+            qkv: torch.Tensor,
+            q_weight: torch.Tensor,
+            k_weight: torch.Tensor,
+        ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+            assert _MINIMAX_QK_NORM_FUSED_OP is not None
+            q_out, k_out = torch.ops.vllm.minimax_qk_norm_fused(
+                qkv,
+                q_weight,
+                k_weight,
+                q_size,
+                kv_size,
+                tp_rank,
+                tp_world,
+                eps,
+                max_tokens,
+            )
+            _, _, v = qkv.split([q_size, kv_size, kv_size], dim=-1)
+            return q_out, k_out, v
+
+        pm.register_replacement(
+            pattern, replacement, self.get_inputs(), pm.fwd_only, pm_pass
+        )
+
+        # Second pattern: three separate split_with_sizes nodes (one per output),
+        # each with _users=1. This occurs when the QKV projection uses a
+        # functional GEMM kernel (e.g. cutlass_scaled_mm via auto_functionalized),
+        # which causes inductor to generate one split per consumer.
+        def pattern_split3(
+            qkv: torch.Tensor,
+            q_weight: torch.Tensor,
+            k_weight: torch.Tensor,
+        ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+            q = qkv.split([q_size, kv_size, kv_size], dim=-1)[0]
+            k = qkv.split([q_size, kv_size, kv_size], dim=-1)[1]
+            v = qkv.split([q_size, kv_size, kv_size], dim=-1)[2]
+            q_fp32 = q.to(torch.float32)
+            k_fp32 = k.to(torch.float32)
+            q_var = q_fp32.pow(2).mean(dim=-1, keepdim=True)
+            k_var = k_fp32.pow(2).mean(dim=-1, keepdim=True)
+            qk_var = torch.cat([q_var, k_var], dim=-1)
+            qk_var = tensor_model_parallel_all_reduce(qk_var) / tp_world
+            q_var, k_var = qk_var.chunk(2, dim=-1)
+            q_out = (q_fp32 * torch.rsqrt(q_var + eps) * q_weight).to(dtype)
+            k_out = (k_fp32 * torch.rsqrt(k_var + eps) * k_weight).to(dtype)
+            return q_out, k_out, v
+
+        pm.register_replacement(
+            pattern_split3, replacement, self.get_inputs(), pm.fwd_only, pm_pass
+        )
+
+
+class MiniMaxQKNormPass(VllmPatternMatcherPass):
+    """
+    Replace forward_qk allreduce+norm with the Lamport fused kernel.
+    Only applied for decode-size compile ranges (small token counts).
+    """
+
+    def __init__(self, config: VllmConfig) -> None:
+        super().__init__(config)
+        self.disabled = True
+
+        if _MINIMAX_QK_NORM_FUSED_OP is None:
+            logger.warning_once(
+                "minimax_allreduce_rms_qk op not found, MiniMaxQKNormPass disabled."
+            )
+            return
+
+        tp_world = get_tensor_model_parallel_world_size()
+        if tp_world <= 1:
+            logger.warning_once("MiniMaxQKNormPass disabled: tp_size <= 1.")
+            return
+
+        if config.model_config is None:
+            logger.warning_once("MiniMaxQKNormPass disabled: no model_config.")
+            return
+
+        hf_cfg = config.model_config.hf_config
+
+        model_name = getattr(hf_cfg, "architectures", "")[0]
+        if model_name != "MiniMaxM2ForCausalLM":
+            return
+
+        num_attention_heads = getattr(hf_cfg, "num_attention_heads", 0)
+        num_key_value_heads = getattr(hf_cfg, "num_key_value_heads", 0)
+        hidden_size = getattr(hf_cfg, "hidden_size", 0)
+        head_dim = getattr(hf_cfg, "head_dim", 0)
+        eps: float = getattr(hf_cfg, "rms_norm_eps", 1e-6)
+
+        if (
+            num_attention_heads != 48
+            or num_key_value_heads != 8
+            or hidden_size != 3072
+            or head_dim != 128
+        ):
+            logger.warning_once(
+                "MiniMaxQKNormPass disabled: cannot infer model info from hf_config."
+            )
+            return
+
+        num_heads_per_rank = num_attention_heads // tp_world
+        num_kv_heads_per_rank = max(1, num_key_value_heads // tp_world)
+        q_size = num_heads_per_rank * head_dim
+        kv_size = num_kv_heads_per_rank * head_dim
+
+        self.max_token_num = min(
+            MAX_TOKEN_NUM, config.scheduler_config.max_num_batched_tokens
+        )
+
+        tp_rank = get_tensor_model_parallel_rank()
+        # Allocate Lamport workspace first.
+        from vllm.distributed.parallel_state import get_tp_group
+        from vllm.model_executor.layers.mamba.lamport_workspace import (
+            get_allreduce_workspace,
+        )
+
+        get_allreduce_workspace(
+            rank=tp_rank,
+            world_size=tp_world,
+            max_tokens=self.max_token_num,
+            process_group=get_tp_group().cpu_group,
+        )
+
+        self.patterns: PatternMatcherPass = PatternMatcherPass(
+            pass_name="minimax_qk_norm_pass"
+        )
+        self._register_patterns(q_size, kv_size, eps, tp_world, tp_rank)
+        self.dump_patterns(config, self.patterns)
+        self.disabled = False
+
+    @enable_fake_mode
+    def _register_patterns(
+        self,
+        q_size: int,
+        kv_size: int,
+        eps: float,
+        tp_world: int,
+        tp_rank: int,
+    ) -> None:
+        MiniMaxQKNormPattern(
+            q_size=q_size,
+            kv_size=kv_size,
+            eps=eps,
+            tp_world=tp_world,
+            tp_rank=tp_rank,
+            max_tokens=self.max_token_num,
+            dtype=self.model_dtype,
+            device=self.device,
+        ).register(self.patterns)
+
+    def is_applicable_for_range(self, compile_range: Range) -> bool:
+        if self.disabled:
+            return False
+
+        return bool(compile_range.end <= self.max_token_num)
+
+    @VllmInductorPass.time_and_log
+    def __call__(self, graph: fx.Graph) -> None:
+        if self.disabled:
+            return
+        self.matched_count = self.patterns.apply(graph)
+        logger.debug("MiniMaxQKNormPass replaced %s patterns", self.matched_count)
+
+    def uuid(self) -> str:
+        return VllmInductorPass.hash_source(self, MiniMaxQKNormPattern)

vllm/compilation/passes/pass_manager.py

@@ -38,6 +38,7 @@ if current_platform.is_cuda_alike():
 if current_platform.is_cuda():
     from .fusion.allreduce_rms_fusion import AllReduceFusionPass
     from .fusion.collective_fusion import AsyncTPPass
+    from .fusion.minimax_qk_norm_fusion import MiniMaxQKNormPass
 
 from .inductor_pass import (
     CustomGraphPass,
@@ -137,6 +138,9 @@ class PostGradPassManager(CustomGraphPass):  # type: ignore[misc]
             if self.pass_config.fuse_allreduce_rms:
                 self.passes += [AllReduceFusionPass(config)]
 
+            if self.pass_config.fuse_minimax_qk_norm:
+                self.passes += [MiniMaxQKNormPass(config)]
+
             if self.pass_config.fuse_norm_quant:
                 self.passes += [RMSNormQuantFusionPass(config)]
                 if rocm_aiter_ops.is_enabled():

vllm/config/compilation.py

@@ -134,6 +134,8 @@ class PassConfig:
     """Enable async TP."""
     fuse_allreduce_rms: bool = None  # type: ignore[assignment]
     """Enable flashinfer allreduce fusion."""
+    fuse_minimax_qk_norm: bool = None  # type: ignore[assignment]
+    """Enable fused allreduce+RMSNorm for MiniMax QK norm."""
     enable_qk_norm_rope_fusion: bool = False
     """Enable fused Q/K RMSNorm + RoPE pass."""
 

vllm/config/vllm.py

@@ -1627,6 +1627,22 @@ class VllmConfig:
                         compile_range_end,
                     )
 
+        if compilation_config.pass_config.fuse_minimax_qk_norm:
+            from vllm.compilation.passes.fusion.minimax_qk_norm_fusion import (
+                MAX_TOKEN_NUM,
+            )
+
+            max_token_num = min(
+                MAX_TOKEN_NUM, self.scheduler_config.max_num_batched_tokens
+            )
+            if compile_range_end is not None and max_token_num < compile_range_end:
+                computed_compile_ranges_endpoints.append(max_token_num)
+            else:
+                logger.debug(
+                    "Max num batched tokens below MiniMax QK norm fusion threshold, "
+                    "MiniMax QK norm fusion enabled for all num_tokens."
+                )
+
         if compilation_config.compile_ranges_endpoints is not None:
             for x in compilation_config.compile_ranges_endpoints:
                 assert isinstance(x, int)

vllm/model_executor/layers/mamba/lamport_workspace.py

+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+
+import array
+import contextlib
+import struct
+import sys
+import threading
+
+import torch
+
+try:
+    from cuda.bindings import runtime as cudart
+except ImportError:
+    from cuda import cudart
+
+_ALIGN = 1 << 21  # 2 MiB — CUDA IPC allocation alignment
+
+
+# ---------------------------------------------------------------------------
+# CUDA helpers
+# ---------------------------------------------------------------------------
+
+
+def _check(error):
+    """Raise on CUDA runtime error."""
+    success = getattr(cudart.cudaError_t, "cudaSuccess", None) or cudart.cudaError_t(0)
+    if error != success:
+        raise RuntimeError(f"CUDA runtime error: {error}")
+
+
+def _cuda_malloc(size: int):
+    aligned = ((size + _ALIGN - 1) >> 21) << 21
+    err, ptr = cudart.cudaMalloc(aligned)
+    _check(err)
+    return ptr, aligned
+
+
+def _cuda_free(ptr: int):
+    if ptr:
+        _check(cudart.cudaFree(ptr)[0])
+
+
+def _cuda_memset_zero(ptr: int, size: int):
+    _check(cudart.cudaMemset(ptr, 0, size)[0])
+
+
+def _cuda_memcpy_d2d(dst: int, src: int, size: int):
+    _check(
+        cudart.cudaMemcpy(
+            dst, src, size, cudart.cudaMemcpyKind.cudaMemcpyDeviceToDevice
+        )[0]
+    )
+
+
+# ---------------------------------------------------------------------------
+# IPC buffer
+# ---------------------------------------------------------------------------
+
+
+class IpcBuffer:
+    """
+    Allocates CUDA device memory and exchanges IPC handles with all ranks
+    so that every rank holds a valid device pointer to every other rank's buffer.
+    """
+
+    def __init__(self, rank: int, world_size: int, size: int, process_group=None):
+        self.rank = rank
+        self.world_size = world_size
+        self.peer_ptrs: list[int] = [0] * world_size
+        self.local_ptr: int = 0
+        self._alive = False
+
+        if size <= 0:
+            return
+
+        self.local_ptr, _ = _cuda_malloc(size)
+        _cuda_memset_zero(self.local_ptr, size)
+        self._alive = True
+
+        # --- exchange IPC handles via torch.distributed ---
+        err, local_handle = cudart.cudaIpcGetMemHandle(self.local_ptr)
+        _check(err)
+
+        all_handles: list[bytes | None] = [None] * world_size
+        torch.distributed.all_gather_object(
+            all_handles, bytes(local_handle.reserved), group=process_group
+        )
+
+        for r in range(world_size):
+            if r == rank:
+                self.peer_ptrs[r] = self.local_ptr
+            else:
+                handle = cudart.cudaIpcMemHandle_t()
+                handle.reserved = all_handles[r]
+                err, ptr = cudart.cudaIpcOpenMemHandle(
+                    handle, cudart.cudaIpcMemLazyEnablePeerAccess
+                )
+                _check(err)
+                self.peer_ptrs[r] = ptr
+
+    def serialize(self) -> list[int]:
+        """Return peer pointers as a list of int64 values (one per rank)."""
+        raw = b""
+        for ptr in self.peer_ptrs:
+            raw += struct.pack("P", ptr)
+        return array.array("Q", raw).tolist()
+
+    def cleanup(self):
+        if not self._alive:
+            return
+        self._alive = False
+        for r in range(self.world_size):
+            if self.peer_ptrs[r] == 0:
+                continue
+            if r == self.rank:
+                _cuda_free(self.peer_ptrs[r])
+            else:
+                with contextlib.suppress(RuntimeError):
+                    _check(cudart.cudaIpcCloseMemHandle(self.peer_ptrs[r])[0])
+            self.peer_ptrs[r] = 0
+        self.local_ptr = 0
+
+    def __del__(self):
+        if not sys.is_finalizing():
+            self.cleanup()
+
+
+# ---------------------------------------------------------------------------
+# Lamport negative-zero initialization
+# ---------------------------------------------------------------------------
+
+
+def _lamport_fill_neg_zero(device_ptr: int, size_bytes: int):
+    """
+    Fill device memory with IEEE-754 negative zero (-0.0f = 0x80000000).
+    This is the "slot empty" sentinel for the Lamport protocol: the kernel
+    spin-waits until a value is *not* negative zero.
+    """
+    if size_bytes == 0 or device_ptr == 0:
+        return
+    n_floats = size_bytes // 4
+    # torch preserves -0.0 in IEEE-754
+    fill = torch.full((n_floats,), -0.0, dtype=torch.float32, device="cuda")
+    _cuda_memcpy_d2d(device_ptr, fill.data_ptr(), size_bytes)
+    del fill
+
+
+# ---------------------------------------------------------------------------
+# LamportWorkspace — the main class
+# ---------------------------------------------------------------------------
+
+
+class LamportWorkspace:
+    """
+    Self-contained workspace for Lamport-based cross-GPU AllReduce.
+
+    Parameters
+    ----------
+    rank : int
+        Local rank (0-based).
+    world_size : int
+        Total number of ranks in the TP group.
+    comm_size : int
+        Size in bytes of *one* Lamport buffer slot. The total IPC allocation
+        per rank is ``3 * comm_size`` (triple-buffering). Must be large enough
+        to hold the per-slot data written by the kernel.  Use
+        ``compute_comm_size_for_minimax()`` for a safe default.
+    process_group : optional
+        ``torch.distributed`` process group for IPC handle exchange.
+        ``None`` uses the default group.
+    """
+
+    def __init__(self, rank: int, world_size: int, comm_size: int, process_group=None):
+        assert world_size >= 2, "Lamport workspace requires at least 2 ranks"
+        assert comm_size > 0, "comm_size must be positive"
+
+        self.rank = rank
+        self.world_size = world_size
+        self.comm_size = comm_size
+
+        # 1) Lamport triple-buffer (the only IPC memory the kernel reads/writes)
+        lamport_total = 3 * comm_size
+        self._lamport = IpcBuffer(rank, world_size, lamport_total, process_group)
+        _lamport_fill_neg_zero(self._lamport.local_ptr, lamport_total)
+
+        # 2) flag_buffer on device: int32[3] = {counter, unused, lamport_flag}
+        #    counter  — used for block-level sync inside the kernel
+        #    unused   — reserved (index 1)
+        #    lamport_flag — triple-buffer rotation index (0 → 1 → 2 → 0 …)
+        self._flag_buf = torch.zeros(3, dtype=torch.int32, device="cuda")
+
+        # 3) layout_buffer on device: int64[2] = {clear_size, comm_size}
+        #    clear_size — bytes to clear from *previous* slot (set by kernel)
+        #    comm_size  — size of one triple-buffer slot
+        self._layout_buf = torch.tensor(
+            [0, comm_size], dtype=torch.int64, device="cuda"
+        )
+
+        # 4) Assemble device-side void* pointer array
+        N = world_size
+        ptrs: list[int] = []
+        ptrs += [0] * N  # [0   .. N-1]   ipc_buffers  (placeholder)
+        ptrs += [0] * N  # [N   .. 2N-1]  ipc_barriers (placeholder)
+        ptrs += self._lamport.serialize()  # [2N  .. 3N-1]  lamport peer ptrs
+        ptrs.append(self._flag_buf.data_ptr())  # [3N]           flag_buffer
+        ptrs.append(self._layout_buf.data_ptr())  # [3N+1]       layout_buffer
+
+        self._workspace = torch.tensor(ptrs, dtype=torch.int64, device="cuda")
+
+    @property
+    def workspace(self) -> torch.Tensor:
+        """Device tensor (int64) that can be passed to the kernel
+        as ``void** workspace``."""
+        return self._workspace
+
+    # ------------------------------------------------------------------
+    # Helpers
+    # ------------------------------------------------------------------
+
+    @staticmethod
+    def compute_comm_size_for_minimax(
+        max_tokens: int,
+        world_size: int,
+        fused_qk: bool = True,
+    ) -> int:
+        """
+        Return a safe ``comm_size`` (in bytes) for MiniMaxReduceRMSKernel.
+
+        The kernel stores per-token variance scalars in the Lamport buffer:
+          - single-matrix path: ``world_size × max_tokens × 4`` bytes per slot
+          - fused Q+K path: ``world_size × 2 × ceil(max_tokens/4) × 16`` bytes per slot
+
+        The returned value is rounded up to 2 MiB alignment.
+        """
+        if fused_qk:
+            groups = (max_tokens + 3) // 4
+            slot_bytes = world_size * 2 * groups * 16  # 16 = sizeof(float4)
+        else:
+            slot_bytes = world_size * max_tokens * 4  # 4  = sizeof(float)
+        return ((slot_bytes + _ALIGN - 1) >> 21) << 21
+
+    def cleanup(self):
+        if hasattr(self, "_lamport"):
+            self._lamport.cleanup()
+
+    def __del__(self):
+        if not sys.is_finalizing():
+            self.cleanup()
+
+    def __repr__(self):
+        return (
+            f"LamportWorkspace(rank={self.rank}, world_size={self.world_size}, "
+            f"comm_size={self.comm_size})"
+        )
+
+
+# ---------------------------------------------------------------------------
+# Cached convenience function (mirrors TRT-LLM's get_allreduce_workspace)
+# ---------------------------------------------------------------------------
+
+_cache_lock = threading.Lock()
+_workspace_cache: dict = {}
+
+
+def get_allreduce_workspace(
+    rank: int,
+    world_size: int,
+    comm_size: int | None = None,
+    max_tokens: int = 16384,
+    process_group=None,
+) -> torch.Tensor:
+    """
+    Return a cached workspace tensor for the given (rank, world_size) pair.
+
+    On first call the workspace is allocated and IPC handles are exchanged;
+    subsequent calls with the same arguments return the cached tensor.
+
+    Parameters
+    ----------
+    rank, world_size : int
+        TP rank and TP size.
+    comm_size : int, optional
+        Explicit slot size in bytes.  If ``None``, computed automatically
+        from ``max_tokens`` and ``world_size`` (fused Q+K path).
+    max_tokens : int
+        Maximum number of tokens per batch (used when ``comm_size is None``).
+    process_group : optional
+        ``torch.distributed`` process group.
+    """
+    if comm_size is None:
+        comm_size = LamportWorkspace.compute_comm_size_for_minimax(
+            max_tokens, world_size, fused_qk=True
+        )
+    pg_id = id(process_group) if process_group is not None else 0
+    key = (rank, world_size, comm_size, pg_id)
+    with _cache_lock:
+        if key not in _workspace_cache:
+            ws = LamportWorkspace(rank, world_size, comm_size, process_group)
+            _workspace_cache[key] = ws
+        return _workspace_cache[key].workspace

vllm/model_executor/models/minimax_m2.py

@@ -233,9 +233,7 @@ class MiniMaxM2Attention(nn.Module):
     ) -> torch.Tensor:
         qkv, _ = self.qkv_proj(hidden_states)
         q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
-        q, k = MiniMaxText01RMSNormTP.forward_qk(
-            self.q_norm, self.k_norm, q.contiguous(), k.contiguous()
-        )
+        q, k = MiniMaxText01RMSNormTP.forward_qk(self.q_norm, self.k_norm, q, k)
         q, k = self.rotary_emb(positions, q, k)
         attn_output = self.attn(q, k, v)
         output, _ = self.o_proj(attn_output)